The availability of geographic information about disease from cancer and birth defect registries has increased public demands for investigation of perceived disease clusters. Many epidemiologists resist this pressure, believing such studies are unproductive and methodologically flawed. Among other criticisms, this view contends that such studies often: combine unrelated diseases; contain too few cases to be meaningful; have "gerrymandered" boundaries; examine only cases without taking into account differences in population density; ignore wide variations in precision among point estimates in various areas; ignore known non-environmental risk factors; ignore latency or other issues regarding timing of exposure relative to disease. While many of these criticisms pertain primarily to neighborhood-level cluster investigations, especially when advanced by non-epidemiologists, some apply to maps made from registry data, e.g., registries typically contain limited data on covariates and record residence at time of diagnosis. On the other hand, well conducted population-based epidemiological studies adhere to design and data collection strictures which make them less subject to criticisms of ad hoc cluster investigations.

We investigated the association between residential history and colorectal, lung and breast cancer on Upper Cape Cod, Massachusetts (USA) using data from two case-control studies of cancer with extensive data on covariates and residential history. We generated maps using generalized additive models (GAMs). Smoothing on latitude and longitude produced maps that estimated the rates of disease relative to the whole study area. By taking into account known risk factors (e.g., age, smoking), we estimated the residual risk due to other potential determinants of disease, including environmental exposures. We used bootstrapping and permutations tests to examine the overall importance of location in the model, construct variance bands and identify local "hot" and "cold" spots. Little or no change was seen in the maps after adjusting for potential confounders, indicating that the individual risk factors that we analyzed did not account for the spatial patterns of disease. No associations between colorectal cancer and location were observed. Both breast cancer and lung cancer "hot spots" tended to increase in magnitude as we increased latency. While we are currently investigating other potential sources of bias, our preliminary results suggests that detailed exposure modeling of certain pollution sources may be warranted.

This file was encoded by the Zend Guard. In order to run it, please install the Zend Optimizer (available without charge), version 3.0.0 or later.

Seeing this message instead of the website you expected?

What is the Zend Optimizer?

The Zend Optimizer is one of the most popular PHP plugins for performance-improvement, and has been available without charge, since the early days of PHP 4. It improves performance by scanning PHP's intermediate code and passing it through multiple Optimization Passes to replace inefficient code patterns with more efficient code blocks. The replaced code blocks perform exactly the same operations as the original code, only faster.

In addition to improving performance, the Zend Optimizer also enables PHP to transparently load files encoded by the Zend Guard.

The Zend Optimizer is a free product available for download from Zend Technologies. Zend Technologies also developed the PHP scripting engine, known as the Zend Engine.

Zend Optimizer not installed

Seeing this message instead of the website you expected?

What is the Zend Optimizer?